From Real Purchase to Realistic Populations of Simulated Customers
Résumé
The use of multiagent-based simulations in marketing is quite recent, but is growing quickly as the result of the ability of such modeling methods to provide not only forecasts, but also a deep understanding of complex interactions that account for purchase decisions. However, the confidence in simulation predictions and explanations is also tightly dependent on the ability of the model to integrate statistical knowledge and the situatedness of a retail store. In this paper, we propose a method for automatically retrieving prototypes of consumer behaviors from statistical measures based on real data (receipts). After preliminary experiments to validate this data mining process, we show how to populate a multiagent simulation with realistic agents, by initializing some of their goals with those prototypes. Endowed with the same overall behavior, validated in earlier experiments, those agents are put into a spatially realistic store. During the simulation, their actual actions reflect the diversity of real customers, and finally their purchase reproduce the original clusters. Besides, we explain how such statistically realistic simulation may be used to support decision in retail, and be extended to other application domains.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...